Detailed view on slow sinusoidal, hemodynamic oscillations on the human brain cortex by Fourier transforming oxy/deoxy hyperspectral images

Abstract Slow sinusoidal, hemodynamic oscillations (SSHOs) around 0.1 Hz are frequently seen in mammalian and human brains. In four patients undergoing epilepsy surgery, subtle but robust fluctuations in oxy‐ and deoxyhemoglobin were detected using hyperspectral imaging of the cortex. These SSHOs were stationary during the entire 4 to 10 min acquisition time. By Fourier filtering the oxy‐ and deoxyhemoglobin time signals with a small bandwidth, SSHOs became visible within localized regions of the brain, with distinctive frequencies and a continuous phase variation within that region. SSHOs of deoxyhemoglobin appeared to have an opposite phase and 11% smaller amplitude with respect to the oxyhemoglobin SSHOs. Although the origin of SSHOs remains unclear, we find indications that the observed SSHOs may embody a local propagating hemodynamic wave with velocities in line with capillary blood velocities, and conceivably related to vasomotion and maintenance of adequate tissue perfusion. Hyperspectral imaging of the human cortex during surgery allow in‐depth characterization of SSHOs, and may give further insight in the nature and potential (clinical) use of SSHOs.

In this article, we provide a thorough description of the observed SSHOs in four patients with epilepsy who underwent epilepsy surgery in the UMC Utrecht. Using Fourier analysis, we were able to produce highly detailed spatial information on the power and phase of oxygenated and deoxygenated hemoglobin SSHOs. These detailed SSHOs analyses may add to a better characterization and understanding of the SSHOs in the human brain.

| Patient characteristics and clinical evaluation for epilepsy surgery
We included four patients (2 males and 2 females, age on average 17 years, 6 4.5 years) who underwent epilepsy surgery between 2009 and 2012. The medical information of the patients is given in Table 1.
Patients 1 and 2 underwent intracranial EEG monitoring during 3-7 days as pre-surgical evaluation. During this clinical monitoring period, electrode grids were placed on the cortex to delineate the seizure onset zone and functional regions. Electrical stimulation mapping (ESM) was applied, in which 30-60 Hz cortical stimulation during 1-7 s was applied to a small area of cortex, while its effect on function was observed. The effect of the stimulation was recorded as either an instantaneous positive effect (e.g., an involuntary contraction of the thumb or any motor behavior), or a subjective experience (e.g., a tingling in a finger). This outcome was used to map functions on the cortex.
In patients 3 and 4, electrode grids were placed on the cortex only during surgery to delineate the epileptogenic region for resection. In patient 3, a Somatosensory Evoked Potential (SSEP) was applied during surgery and the central sulcus was located. With ESM, motor and sensory leg was located interhemispherically. No motor mapping was performed on visible cortex, so no motor and sensory hand/arm were located. In patient 4, no SSEP or ESM was performed.

| Hyperspectral imaging
In all patients, hyperspectral recordings were made during surgery, while the surgeon was not operating. The recordings were approved by the Medical Ethical committee of the University Medical Center of Utrecht. All four patients gave informed consent to make hyperspectral recordings during surgery.

| Hyperspectral imaging system 1
Patient 1 was imaged in 2009 during grid explanation surgery, with a hyperspectral camera system to see whether changes in oxygenation could be perceived by the imaging system. This system consisted of a hyperspectral camera mounted to the assistant's tube of a surgical microscope ( Figure 1a). It consisted of a liquid crystal tunable filter (Cri VariSpec VIS bandwidth (FWHM) 10 nm) and a monochrome camera with 1392 3 1024 resolution (PCO PixelFly QE) . A polarizer was placed in the incident path of the microscope light, cross-polarized to the tunable filter, to suppress reflections from the surface.
The patient was continuously imaged for 7 min looping over 4 specific spectral bands (duration 0.46 s per spectral band). The imaging at this group of 4 spectral bands was repeated 200 times. The central wavelengths of the spectral bands were 480, 570, 600, and 660 nm, which were chosen such that the oxy-and deoxy-absorption spectrum differed at these wavelengths and that the light was scattered at deeper tissue layers while still within the transmission range of the Cri filter. With this system oxygenation values could be determined once per 4 3 0.46 s 5 1.8 s.
During the measurements, a seizure was captured. Measurements showed increase in blood flow during the seizure . In this study, we only included the interictal measurements.

| Hyperspectral imaging system 2
A second version hyperspectral system with hyperspectral light source and monochrome camera was used for patients 2,3, and 4 in 2012, in which detection of oxygenation values was four times faster, roughly 2 times per second.
This system consisted of a flat panel light source with a total of 600 LEDs with 17 different peak wavelengths (of which 7 were used in these experiments), with a CMOS camera mounted in the middle of the panel (BCi5 CMOS Camera, C-Cam Technologies) ( Figure 1b) (Klaessens, Nelisse, Verdaasdonk, & Noordmans, 2013)  suppressed (Supporting Information Figure 1). As more wavelengths were captured, the absolute oxy-and deoxyhemoglobin concentrations could be calculated and, thus, the saturation values.

| Acquisition parameters
The acquisition parameters for all patients are given in Table 2. It shows that with the second system, we could image faster and image more wavelengths within the same period of time. The increasing exposure time of the camera for patients 2 to 4 reflects increased optimization of the exposure time for a good signal while not overflowing the 8-bit CMOS camera.

| Post-processing
For both systems, reflection (backscatter) images were calculated by dividing the acquired images by the spectrum from a white surgical tissue placed on the brain (visible during acquisition). Motion (translation and rotation) was compensated using custom written image registration software (Noordmans, van den Biesen, de Roode, & Verdaasdonk, 2008). The oxygenation and de-oxygenation values were calculated by modeling the spectral absorption of the light traveling through the top layers of the brain tissue and scattered back to the camera. This was done by two different calculation methods, the Delta-t method and the fit-method (Klaessens et al., 2013) (See also Supporting Information Figure 2): Delta-t method: For all patients, we used the delta-t method to get a strong signal of the changes in oxygenation Dc HbO and Dc HbR . Using the reflection image Rx; t; k ð Þ, (wherex denotes position, t time, and k wavelength) at t50 s as a reference, the Delta-t formula becomes: (1) where A HbO k ð Þ and A HbR k ð Þ represent the absorption spectra of oxyhemoglobin and deoxy-hemoglobin respectively. The change in concentrations were then calculated using least square fitting with pseudo    Figure 6E, near retractor). (b) Corresponding power spectra. Clearly peaks are seen for the SSHOs around 0.1 Hz, and the respiration and heart rate, the latter two corresponding to the rates reported by the anesthesia machine of 12 and 56 min 21 , respectively. (c) Zooming on the curves of (a). A phase shift of about 180 degrees can clearly be seen between the changes in concentration of oxygenation and deoxygenation [Color figure can be viewed at wileyonlinelibrary.com] into an oxygenation concentration c HbO , a deoxygenation concentration c HbR , and two (wavelength independent) constants C 1x ; t ð Þ and C 2x ; t ð Þ.
The concentrations were then calculated using least square fitting with pseudo inverse matrix calculations. The total hemoglobin is then calculated by and the saturation by The Fit-method results in a set of c HbO and c HbR images, where the HbO and HbR concentrations are known per pixel at each moment in time.

| Frequency analysis
To analyze the SSHOs, which were exceptionally stationary during the The reason to choose for the Delta-t HbO signal here was that the Delta-t method yielded a stronger signal than the Fit-method and that the HbO signal appeared to be stronger than the HbR signal. These regions were manually segmented using a custom-made drawing program written in Java (Java navigation), and subsequently numbered from lower to higher SSHO frequencies. For each SSHO-region the following parameters were extracted: FIG URE 4 (a) Photo, amplitude and Du region; DHbO phase images of four SSHO regions of patient 1. The images are intensity stretched individually to enhance details. See gray scales for exact values which can also be read from Table 3. Arteries do or just do not take part in the SSHO. The maximal phase difference within a region can approach one period, thus tens of seconds apart. (b) Amplitude and D u region;D HbO phase images of four SSHO regions of patient 1 within the same area. The images are intensity stretched individually to enhance details. See gray scales for exact values which can also be read from

| Heart rhythm and respiration
For all patients, faster oscillations like respiration and sometimes heart rhythm were observed. The respiration rate of 12 to 18 min 21 corresponded to that supplied by the anesthesia machine.
The heart rate (around 56 min 21 ) was only observed in patient 3 where the sampling rate of the hyperspectral camera was sufficiently high above the Nyquist criterion. This heart rate corresponded with the heart rate recorded by the anesthesia machine ( Figure 2). Both the respiration and heartrate effects were seen across the entire exposed cortex, contrary to the local character of the SSHOs described next.

| Location of SSHO regions
In patient 1, 25 distinct regions were seen with different SSHO frequencies ( Figure 3, Table 3), ranging from 0.044 Hz to 0.082 Hz.
Region 1 is omitted as it is actually not a SSHO, but corresponds to the region with increased perfusion only once during the image acquisition . Some regions did not overlap like region 6,  Table 4). The SSHO's amplitude was 4.5 mM, which was in the same range as for patient 1. The change in saturation was 1.8 percentage points (pp) at a mean saturation value of 56%. The saturation in other parts of the exposed brain without SSHO-regions was a bit higher: 70% on the average. With ESM, one area was found with both motor and sensory hand. This area was located in the parietal cortex and overlapped exactly with the SSHO-region that was found with the spectral camera. from 0.8 pp to 2.2 pp (region 9). The saturation in the SSHO-regions was 61% on the average, and 63% in the brain without SSHO-regions.
In this patient, no ESM was performed. Based on the location of central sulcus, part of the SSHO-region seemed to be anatomically located on the sensory hand area (parietal cortex).
The observed SSHO phase difference within a region could be explained by a propagating hemodynamic wave. The wavefront velocity statistics for all regions and patients are shown in Figure 9. The SSHO wavefront velocity across all patients was found to be: average value 0.8 mm/s (0.4, 0.8, 1.3, 0.7 mm/s respectively for patients 1-4) and median value 0.5 mm/s (0.3, 0.5, 0.9, 0.5 mm/s).

| D ISC USSION
SSHOs were observed in the oxy-and deoxyhemoglobin concentration variations on the cortical surface obtained with intraoperative hyperspectral imaging in four patients with epilepsy. We observed SSHO-regions with different specific frequencies that often showed spatial overlap. The phase shift between oxy-and deoxyhemoglobin SSHOs D/ DHbO2DHbR was about half an oscillation period, i.e. almost 1808. On average, the amplitude of the deoxyhemoglobin SSHOs was 11% smaller than that of oxyhemoglobin SSHOs. The maximal phase variation within a SSHO-region

| F I ND I NG S
The major findings in this article are:

| Oscillating frequencies are stationary over time and have a very small bandwidth
Instead of applying a relatively wide bandpass filter, where all oscillations are aggregated (Bumstead et al., 2016;Gratton et al., 1998;Zuo et al.,   In this way, it becomes clear that different but also overlapping SSHOregions can be observed with slightly different SSHO frequencies.

| Oscillating frequencies differ between patients
While for patient 2 to 4 the oscillating frequencies lay between 0.08 and 0.12 Hz, the frequencies for patient 1 were lower, between 0.04 and 0.08 Hz. Retrospectively, technical tests were performed on software from 2009 and 2012 to confirm that the timings were correct.
Anaesthetics appeared not to be that different that it could explain the difference in oscillation frequency. Differences in pathology, stress and location of resection could play a role, but we could not find such clues why the SSHO frequencies for patient 1 differed almost a factor two from the SSHO frequencies of patient 2 to 4.

| SSHOs are found in local regions
It is striking that the SSHO-regions are localized to specific anatomical areas. For patient 1, 2 and 3 the regions are located near the motor and somatosensory strip, for patient 4 around the more posterior part of the parietal area. Why the SSHO-regions were only found there, remains unclear.
In patient 4, one SSHO-region is located on the parietal cortex, outside the somatosensory area. The other three patients only showed SSHO-regions in the somatosensory or motor area. The more posterior parietal cortex was not visible. Therefore, we do not know whether SSHO-regions would also have been present in the parietal cortex in patient 1-3. The observation of the spectral camera should be repeated in more patients with visible parietal cortex during surgery.

| F UTUR E DI R EC TIONS
Although our analysis gives new insight in the characteristics of SSHOs (distinct regions and various stable frequencies), several questions remain to be answered: 1. It has been hypothesized that SSHOs are linked to pathology (Rayshubskiy et al., 2014) and that tissue in stress can produce SSHOs.
We think that stress may indeed play a role, but it is a factor that is difficult to measure. Mechanical stress could play a role, but if it would, one would expect to see SSHOs mainly at the edges of the skull opening, or more widespread over the entire exposed brain, while we see very focal regions. There might be a relation with the lower saturation values found within the SSHO-regions. This was certainly the case for patient 2 and 3, but for patient 4 the saturation within the SSHOregions was only slightly smaller than outside these regions.
However, it seems doubtful that the brain is 'wasting energy' in producing such oscillations (Fox & Raichle, 2007). SSHOs may rather be related to active functional areas where perfusion is critical and SSHOs reflect vasomotion, generated to improve perfusion (C. Aalkjaer et al., 2011). By avoiding a 'latch state' the small vessels are ready to increase their performance in perfusing the brain tissue. The brain tissue is very active, and it is important that the blood perfusion is ready to improve its capacity.
3. Several models have been proposed to describe hemodynamic changes in terms of arterial, capillary and venous compartments.
Although they explain the BOLD response (Boas, Jones, Devor, Huppert, & Dale, 2008) and (slow) oscillations at a coarse level (Fantini, 2014), they do not explain the characteristics of the slow oscillations we found in our measurements. Our measurements confirm the phase difference of 1 p between the oxy and deoxy oscillation found by others (Reinhard et al., 2006;Watanabe et al., 2017). These models lack an explanation for the local (sometimes overlapping) character of the slow oscilla- In the analysis of resting state functional MRI, correlations are found in the BOLD signal between brain areas across the entire brain (Fox & Raichle, 2007), producing the so called resting-state networks such as the default mode network. Knowing that SSHOs exist, having very steady frequency and a local character, one might model and include this in the analysis to identify resting state networks within FIG URE 9 Wavefront velocity statistics for each SSHO based on Im Phase DHbO phase images. For each region, the minimum velocity is given by the lowest point of the lowest line, the median velocity by the box bottom, the average velocity by the box top and the maximum velocity by the highest point of the highest line (clipped to zoom in on average and median velocity). Note that the wavefront velocity of patient 1 was almost a factor 2 lower than for the other patients brain regions, i.e. at a much smaller spatial scale than previously studied in resting-state fMRI.

| Scattering
Note that scattering has been ignored in these calculations. We acknowledge that it should be included for correct estimation of hemoglobin concentrations. However, it is unclear how scattering should be included for reflection (backscatter) imaging of the human brain. Most values are derived from bloodless ex-vivo brain tissue, while blood will be the major absorber and prevent light to be scattered and thus reach the penetration depth as specified by literature (Sterenborg, Gemert, Kamphorst, Wolbers, & Hogervorst, 1989). To get an idea on possible impact, we recalculated the hemoglobin concentration when assuming a three times higher penetration depth at 850 nm compared to 470 nm. The hemoglobin concentration then became 40% lower. Furthermore, this will only affect the absolute concentration and saturation values. It does not affect the amplitudes of the HbO, HbR and thus saturation SSHOs.

| Region selection
The selection of the SSHO-regions by hand can be considered arbitrary. Although the strongest oscillations were easy to recognize, there were often regions that oscillated weakly. It is hard to devise an algorithm to automatically extract the regions as the boundaries are difficult to define when the oscillating amplitude approaches that of the 'noise level'. Also, larger vessels produce artefacts (probably due to motion) having strong power at the oscillating frequency, preventing the use of an easy thresholding algorithm.

| Time analysis
As the SSHOs appeared to be exceptionally stationary, Fourier analysis could be used to determine the amplitude and phase of the sinusoidal oscillations. When this would not be the case, e.g. when the SSHOs vanish and reappear, more advanced methods like wavelet, spectrogram or Hilbert transforms might be needed to study this dynamic behaviour (Watanabe et al., 2017).

| C ONC LUSI ON S
SSHOs in oxygenation and deoxygenation could be found in all 4 patients with epilepsy. As the SSHOs are very stable over time, a Fourier transform can be used to study spatial amplitude and phase patterns of these SSHOs. It appears that many SSHO-regions with slightly different frequencies can be seen on the cortex. Often SSHO-regions do overlap giving a clear view on the individual vascular networks perfusing brain tissue. Hyperspectral imaging of the human cortex during surgery allow in-depth characterization of SSHOs, and may give further insight in the nature and potential (clinical) use of SSHOs.